Mahmoud Khalil

Experience

• 

  Mathnasium - The Math Learning Center

  Oct 2017 - Jan 2018

  Math Instructor

  • Conducted one-on-one instruction with students, accommodating up to three students simultaneously.• Developed individualized learning plans tailored to the unique needs and learning styles of each student.

• 

  Khalifa University

  Jan 2018 - Dec 2019

  Machine Learning Research Assistant

  • Assembled a real light curve time-series dataset containing 16,193 entries of space objects, distinguishing it from available simulated datasets.• Analyzed the dataset, conducted preprocessing, and extracted 53 meaningful features using FATS statistical library in Python.• Developed, optimized, and evaluated ML models using TensorFlow for deep learning tasks (ANN, CNN), and scikit-learn for traditional ML algorithms (SVM, k-NN, Random Forest and decision trees) for space debris classification.• Reduced feature dimensionality by 80%, and applied oversampling for dataset balance, enhancing model precision by 74%.• Results published in 2 conference papers. Show less

• 

  Ministry of Education – UAE

  Mar 2020 - Dec 2023

  Computer and Mathematics Instructor

  • As PLC coordinator, led 11 teachers in collaborative discussions, fostering best practices and enhancing student outcomes.• Conducted statistical analysis on students' marks and performance data to identify areas of improvement.• Achieved an 8% increase in students' average scores on MAP standardized test in Fall 2023 compared to the previous academic year.• Taught Python, machine learning, NLP, computer vision, algebra, probabilities, and statistics.• Received a 94% positive feedback rating from students. Show less

• 

  Khalifa University

  Jan 2024 - Aug 2024

  Machine Learning Engineer

  • Tested and evaluated 6 prompt engineering techniques, including zero-shot, few-shots, chain-of-thought prompting, and using LangChain and LlamaIndex packages. Compared model evaluations to human expert evaluations of medical content quality, resulting in a 15% improvement in model response quality.• Employed 16 language models (LLMs) including GPT-4, Claude 3, Gemini (via APIs), Llama 3, Mistral, Phi-3, and Falcon (via Transformers library) to assess transcript credibility based on established medical standards (DISCERN instrument).• Leveraged Batch API service for GPT-4, and GPUs with CUDA while employing Transformers models for faster computation.• Curated 458 medical educational videos from YouTube.• Implemented OpenAI speech-to-text system for accurate transcription, and wrote a Node.js script to fetch transcripts from YouTube.• Utilized NLTK and Pandas for Natural Language Processing (NLP) analysis and data filtering:- Excluded non-English and irrelevant transcripts.- Calculated similarity metrics between OpenAI and YouTube transcripts, selecting the best transcripts from each.- Performed supplementary analysis: identified most repeated words and phrases, tokenized transcripts, and generated histograms. Show less